Does Java 8 have cached support for suppliers?

Question

The guava library has it's own Supplier which does not extend Java 8 Supplier. Also guava provides a cache for suppliers - Suppliers#memoize.

Is there something similar, but for Java 8 Suppliers?

Answer 1

There's no built-in Java function for memoization, though it's not very hard to implement it, for example, like this:

public static <T> Supplier<T> memoize(Supplier<T> delegate) {
    AtomicReference<T> value = new AtomicReference<>();
    return () -> {
        T val = value.get();
        if (val == null) {
            val = value.updateAndGet(cur -> cur == null ? 
                    Objects.requireNonNull(delegate.get()) : cur);
        }
        return val;
    };
}

Note that different implementation approaches exist. The above implementation may call the delegate several times if the memoized supplier requested simultaneously several times from the different threads. Sometimes such implementation is preferred over the explicit synchronization with lock. If lock is preferred, then DCL could be used:

public static <T> Supplier<T> memoizeLock(Supplier<T> delegate) {
    AtomicReference<T> value = new AtomicReference<>();
    return () -> {
        T val = value.get();
        if (val == null) {
            synchronized(value) {
                val = value.get();
                if (val == null) {
                    val = Objects.requireNonNull(delegate.get());
                    value.set(val);
                }
            }
        }
        return val;
    };
}

Also note, as @LouisWasserman correctly mentioned in comments, you can easily transform JDK supplier into Guava supplier and vice versa using method reference:

java.util.function.Supplier<String> jdkSupplier = () -> "test";
com.google.common.base.Supplier<String> guavaSupplier = jdkSupplier::get;
java.util.function.Supplier<String> jdkSupplierBack = guavaSupplier::get;

So it's not a big problem to switch between Guava and JDK functions.

Answer 2

The simplest solution would be

public static <T> Supplier<T> memoize(Supplier<T> original) {
    ConcurrentHashMap<Object, T> store=new ConcurrentHashMap<>();
    return ()->store.computeIfAbsent("dummy", key->original.get());
}

However, the simplest is not always the most efficient.

If you want a clean and efficient solution, resorting to an anonymous inner class to hold the mutable state will pay off:

public static <T> Supplier<T> memoize1(Supplier<T> original) {
    return new Supplier<T>() {
        Supplier<T> delegate = this::firstTime;
        boolean initialized;
        public T get() {
            return delegate.get();
        }
        private synchronized T firstTime() {
            if(!initialized) {
                T value=original.get();
                delegate=() -> value;
                initialized=true;
            }
            return delegate.get();
        }
    };
}

This uses a delegate supplier which will do the first time operation and afterwards, replace itself with a supplier that unconditionally returns the captured result of the first evaluation. Since it has final fields semantics, it can be unconditionally returned without any additional synchronization.

Inside the synchronized method firstTime(), there is still an initialized flag needed because in case of concurrent access during initialization, multiple threads may wait at the method’s entry before the delegate has been replaced. Hence, these threads need to detect that the initialization has been done already. All subsequent accesses will read the new delegate supplier and get the value quickly.

Answer 3

A simple wrapper for Guava 20 on Java 8:

static <T> java.util.function.Supplier<T> memoize(java.util.function.Supplier<? extends T> supplier) {
    return com.google.common.base.Suppliers.memoize(supplier::get)::get;
}